In the Fraser River mouth region of British Columbia, commercial fishing of sockeye salmon by Canadian and American enterprises occurs within a complex system of straits between Vancouver Island and mainland Canada. Fishing seasons and the catch ratio of sockeye salmon between the USA and Canada depend on the characteristics of their return migration through these straits. This study posits that sockeye salmon behavior is influenced by seawater chemistry, sea surface temperature, and current dynamics. Training sample was created using various open reanalysis data variables, including local currents, surface current velocities, salinity, sea surface temperature, wind speed, wave characteristics, sea level pressure, air temperature, humidity, and precipitation. Classical machine learning models, such as linear regression, Ridge regression, and Random Forest, were employed to predict the northern diversion rate (NDR) and median return timing dates for salmon migration to the Fraser River estuary. The training dataset was sourced from Glorys2v4 reanalysis data, which describes ocean dynamics. The findings suggest areas for improvement, such as optimizing control points through probabilistic stochastic or gradient ascent optimization and employing neural network techniques like MoCo and ConvLSTM for data assimilation to mitigate data gaps. The results underscore the challenges and opportunities for enhancing the prediction quality of salmon migration using machine learning models and reanalysis data.